Statistical atlases, and associated image analysis methods, have found widespread use in several neuroimaging fields, presenting a powerful way to integrate diverse imaging information, correlate it with genetic and clinical measurements, understand effects of disease on brain structure and function, and construct diagnostic tools. This proposal will combine statistical image analysis, deformable registration, and biophysical modeling approaches to an integrated framework for constructing and clinically using statistical atlases from brain tumor patients. Emphasis is placed on gliomas, which have very poor prognosis due to cancer infiltration beyond the visible tumor boundary. Accordingly, the ultimate clinical goal of this study is to identify subtle imaging characteristics of brain tissue that is likely to be infiltrated by tumor, as well as of tissue that is likely to present recurrence in relatively shorter time period. This will be achieved by studying the multi-modal imaging phenotypes of healthy and pathologic tissues in conjunction with spatial information, including the spatial pattern of the tumor and the proximity of malignant tissue to white matter fiber pathways, and by correlating these phenotypes with clinical information, including tumor recurrence. The hypothesis is that signal and spatial information together will be able to identify brain tissues that are likely to later present recurrence. The main technical challenges that will be overcome are 1) development of computationally efficient biophysical models of tumor growth, diffusion, and mass effect; 2) development of deformable registration methods that will allow us to co-register tumor-bearing brain images and build a population-based atlas-the main challenges here are to estimate the appropriate tumor parameters as well as the location of peri-tumor anatomy that is typically confounded by edema, infiltration and extreme deformations; and 3) development of machine learning methods for characterizing subtle abnormalities of brain tissue, and for identifying tissue that is likely to present recurrence after resection and treatment. Pilot studies on the feasibility of this approach to larger clinical studies will be performed on a database of brain MR images obtained from glioma patients via a rich and extensive acquisition protocol, including perfusion, diffusion tensor imaging, spectroscopy, and conventional imaging.